System Prompt Security — Protect Against Extraction & Injection

I will harden your system prompt against security vulnerabilities — testing for prompt injection attack susceptibility, system prompt extraction through adversarial queries, persona override attempts, jailbreak patterns, and indirect injection through tool outputs or user-provided documents, with specific hardening modifications for each vulnerability found. System prompts that define application behaviour are frequent attack targets — users attempt to extract the full system prompt to reverse-engineer your application, inject instructions that override your constraints, or manipulate the AI into behaviours you explicitly prohibited; hardening requires both testing to discover vulnerabilities and prompt modifications that structurally resist the attack patterns.

The hardening covers injection resistance testing across 20+ attack patterns, extraction resistance testing, persona override attempts, jailbreak pattern testing, indirect injection via document inputs, and prompt modifications addressing each vulnerability with post-modification retesting.